Hooks

Hooks let you automate reactions to events in Crew. When a tool fires, a session starts, or a file changes, a hook can run a script, inject context, or gate the action.

Open Agent Capabilities → Hooks. You can view, create, edit, and delete hooks from the browser. Each hook shows its trigger, action, and optional matcher pattern.

Hooks are JSON files at ~/.kiro/crew/hooks/ (or project-level at .kiro/hooks/). Each file can define one or more hooks.

A hook has three parts:

1. Trigger — when does it fire?
2. Matcher (optional) — which specific events match? (regex on tool name or file path)
3. Action — what happens? (run a command, or inject a prompt)

Runs a shell command. The command receives session context as JSON on stdin.

Use for: running linters, triggering builds, sending notifications, validating changes.

Exit codes matter:

• 0 — success (output forwarded for some triggers)
• 2 — block the action (for PreToolUse, UserPromptSubmit)
• Other — silent failure, no block

Injects a static prompt into the model context.

Use for: adding reminders, enforcing patterns, injecting context about the current state.

A hook that runs the linter whenever a TypeScript file is saved:

• Trigger: PostFileSave
• Matcher: \.(ts|tsx)$
• Action: command — npm run lint

A hook that adds a reminder before any file write:

• Trigger: PreToolUse
• Matcher: fs_write|str_replace
• Action: agent — "Verify this write follows our coding standards before proceeding."

• Trigger: PostTaskExec
• Action: command — npm run test

PreToolUse hooks can block a tool call by exiting with code 2. This is how you add custom deny rules beyond the built-in 137 patterns. The agent sees the block and tries a different approach.

1. Open Agent Capabilities → Hooks
2. Click Add hook
3. Choose a trigger, optionally set a matcher regex, select the action type
4. For command actions: enter the shell command
5. For agent actions: enter the prompt text
6. Save

The hook is active immediately — no restart needed.